Note: This blog post was originally written in Japanese for our Japanese website. We used our machine translation platforms to translate it and post-edit the content in English. The original Japanese post can be found here.
Some of you may not be familiar with the term "post-editing," but it has become one of the major methods in the translation industry. As a translation agency that works on post-editing projects on a daily basis, we’ll introduce what post-editing is, the differences from human translation, and the types of post-editing in this post.
In the translation and localization world, many abbreviations related to machine translation are used. Here are some of them.
|
Types of post-editing: full PE and light PE
As explained above, post-editing refers to completing a translation by editing the output of machine translation (MT). Depending on the required quality, post-editing (PE) is generally divided into two types: full PE (defined by ISO 18587) and light PE.
Simply put, full PE is almost the same as human translation in terms of quality requirements, except that it utilizes as much of the MT output as possible.
Light PE does not require as high a quality as full PE or human translation; the main goal is that the meaning is conveyed. Depending on the client requirements, there may be cases where little to almost no editing is done to the MT output.
In summary, the following applies.
|
Full PE (defined by ISO 18587): By leveraging the output of MT, it is possible to achieve higher productivity than human translation while still meeting the basic quality requirements similar to human translation. |
*For Kawamura International’s post-editing services, we do not classify into full PE or light PE, but instead take a more flexible approach by proposing the optimal post-editing plan according to our client's needs.
What’s the difference between PE and human translation?
The most significant difference between post-editing and traditional human-only translation is the process. In human translation, the translator translates the original text, and a separate reviewer checks the translation against the original text (in some cases, this check may not be performed depending on the client requirements). In contrast, in post-editing, the original text is first machine-translated and then corrected by a human.
As you may have noticed from the description of full PE, post-editing requires the same level of translation skills as translating from scratch. This is because it’s necessary to compare the original text and the MT output and make corrections.
Post-editing is often thought to be several times faster than human translation because it uses MT, but as mentioned above, the time required varies depending on the expected quality. For example, if you prioritize cost reduction and speed over quality, light PE is suitable, as it focuses on only corrections of errors and missing information.
The evolution of MT
In the past, there were several types of MT that preceded post-editing, such as statistical machine translation (SMT) and neural machine translation (NMT).
SMT is a method that generates translations from large amounts of bilingual data using statistical techniques. It’s one of the conventional MT methods, but it sometimes produced translations that made people wonder, "Why does entering this sentence result in such a translation?"
In contrast, NMT, which has become mainstream in recent years, is a type of MT that uses artificial intelligence (AI) to learn correct sentences and improve translation accuracy. Sometimes, it produces high-quality translations that are hard to believe were generated by a machine.
And now, Generative AI (GenAI) is also in the mix. It can produce more human-like translations in terms of readability, especially in nuanced or complex situations.
Without such advances in MT, it would not be an exaggeration to say that the demand for post-editing would not have increased as much as it has today. Post-editing of conventional MT, such as SMT, required almost the same amount of effort as human translation, but this is now being greatly improved by the ongoing evolution of NMT and GenAI.
Human vs. machine
What is it that AI can never truly learn? It is human qualities such as emotions and culture.
This is another important point when considering the differences between human translation, MT, and post-editing. While AI can learn correct sentences, it cannot absorb emotions, culture, or context. As a result, MT outputs before post-editing inevitably contain some literal translations.
On the other hand, in human translation, human errors can occur due to carelessness, such as including information not present in the original, omitting information that is present, or misreading information like numbers.
It’s possible to pre-edit the original text to make it easier for MT to improve its output, but whether this is suitable still depends on the content of the original text. Since it’s impossible to do away with either MT or human translation, the ideal situation is for both to complement each other.
Post-editing used to be seen as a task that many translators tend to avoid. Some of the commonly heard comments were: “Seeing unnatural sentences affects my own writing to also be awkward,” “The pay is lower than for general translation work,” and “My translation skills are not fully utilized.”
Post-editing certainly requires skills that differ from those needed when translating from scratch (such as making the most of MT output when revising, and efficiently identifying and checking for common MT errors, among others). However, the required level of translation expertise remains unchanged, and it’s necessary to make use of the knowledge and skills that are cultivated through translation work.
Now, with the improvements in machine translation and corpora increasing the accuracy of MT outputs, the burden of post-editing has reduced, and these views seem to be changing. While the demand for human translation will never disappear, the demand for post-editing will surely continue to grow.
MT once required a lot of effort to correct, but now, in some cases, it can be used without editing the output at all. As MT has evolved, post-editing has evolved as well.
Unlike human translation, where a translator works from scratch, post-editing is the process of revising the output of MT. By utilizing post-editing together with MT, you can reduce cost, improve translation efficiency, and speed up the process.
For example, you can use different approaches of post-editing in the following ways:
|
Demand is rising for translation services using post-editing (PE) that makes the most of MT output.
At Kawamura International, our experienced staff will propose the best solution depending on your priorities, such as quality, delivery time, or cost, to meet your challenges and needs. If you have any questions about translation or localization, please feel free to reach out to us.